When to Use Encode-Prefill-Decode Disaggregation to Accelerate Multimodal Model Serving

<img width="768" height="432" src="https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-768x432.jpg" class="webfeedsFeaturedVisual wp-post-image" alt="" style="display: block; margin-bottom: 5px; clear:both;max-width: 100%;" link_thumbnail="" decoding="async" srcset="https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-768x432.jpg 768w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-179x101.jpg 179w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-300x169.jpg 300w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-625x352.jpg 625w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-1536x864.jpg 1536w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-645x363.jpg 645w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-660x370.jpg 660w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-500x281.jpg 500w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-160x90.jpg 160w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-362x204.jpg 362w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-195x110.jpg 195w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-1024x576.jpg 1024w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-960x540.jpg 960w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo.webp 1999w" sizes="(max-width: 768px) 100vw, 768px" title="multimodal-dynamo" />Encode-prefill-decode (EPD) disaggregation is an inference optimization technique for multimodal models that separates the vision encoder stage from the prefill...<img width="768" height="432" src="https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-768x432.jpg" class="webfeedsFeaturedVisual wp-post-image" alt="" style="display: block; margin-bottom: 5px; clear:both;max-width: 100%;" link_thumbnail="" decoding="async" loading="lazy" srcset="https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-768x432.jpg 768w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-179x101.jpg 179w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-300x169.jpg 300w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-625x352.jpg 625w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-1536x864.jpg 1536w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-645x363.jpg 645w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-660x370.jpg 660w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-500x281.jpg 500w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-160x90.jpg 160w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-362x204.jpg 362w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-195x110.jpg 195w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-1024x576.jpg 1024w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo-960x540.jpg 960w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/multimodal-dynamo.webp 1999w" sizes="auto, (max-width: 768px) 100vw, 768px" title="multimodal-dynamo" /><p>Encode-prefill-decode (EPD) disaggregation is an inference optimization technique for multimodal models that separates the vision encoder stage from the prefill and decode stages. It is most effective for image-heavy prompts, short-to-medium outputs, and quantized mixture-of-experts (MoE) models. This post shows when and how to use EPD disaggregation with NVIDIA Dynamo to achieve up to 5x…</p> <p><a href="https://developer.nvidia.com/blog/when-to-use-encode-prefill-decode-disaggregation-to-accelerate-multimodal-model-serving/" rel="nofollow" data-wpel-link="internal" target="_self">Source</a></p>